Bayesian Decision Trees for EEG Assessment of newborn brain maturity

Abstract

Decision Tree (DT) models are observable for clinical experts and can be used for a probabilistic inference within Bayesian Model Averaging (BMA). The use of Markov Chain Monte Carlo (MCMC) search makes the BMA computationally practical. We employ the MCMC BMA strategy for assessing newborn brain maturity from clinical EEG. Our analysis has revealed that an appreciable part of EEG features is rarely used in the DT models, because these features make weak contribution to the assessment. It was also found that the portion of DT models using weak EEG features was large. On one side, this obstructs interpretation of DT models. On the other side, weak attributes increase dimensionality of a model parameter space that MCMC needs to explore in detail. We assume that discarding the DT models using weak features will reduce these negative impacts. Specifically, in this paper we explore the influence of pruning DTs on the results obtained within the discarding technique we proposed. Our experiments have shown that, given a pruning factor, the original set of EEG features can be greatly reduced without a decrease in accuracy of assessment.